Common Failure Modes at Scale

A scaled agent system can fail in several predictable ways.

The session contains too many rules, tools, outputs, and historical messages. Quality declines and compaction removes important detail.

A local preference becomes a global rule and damages unrelated work.

Every workflow is loaded in full even when most skills are irrelevant.

A weak model receives a task requiring difficult reasoning, or an expensive model receives a simple repetitive task.

Many agents are spawned without a clear benefit, multiplying cost.

Independent agents return useful work, but the parent fails to compare or integrate it.

The creator is asked to approve its own work after accumulating a large biased context.

The system assumes automatic compression will preserve every critical detail.

Agents continue acting because no definition of done exists.

The architecture is technically successful but economically unsustainable.

Each failure is addressed by one or more of the design patterns in this module.